Generative AI for Phishing Detection

Applied Large Language Models (LLMs) to generate phishing emails and analyze their impact on cybersecurity defenses, providing insights for AI-driven security strategies.

Motivation

AI-generated phishing emails are increasingly sophisticated. This project aims to study these threats and evaluate detection methods to strengthen organizational security posture.

Tools & Technologies

  • LangChain & Ollama for LLM integration
  • Python for data processing and analysis
  • Machine learning models (supervised classifiers)
  • Visualization with matplotlib and seaborn

Methodology

  • Generated phishing emails using LLMs with diverse personas to increase dataset variety.
  • Created labeled datasets of AI-generated phishing emails.
  • Trained supervised learning models to detect phishing content.
  • Evaluated models using accuracy, precision, recall, and F1-score metrics.

Results & Key Takeaways

Enhanced detection of AI-generated phishing threats, showcasing the importance of LLMs in cybersecurity research. Learned methods to diversify training data and improve model performance using personas.